Research on Improved YOLOv10n Sign Language Recognition Algorithm

In diverse social communication scenarios, sign language is a crucial communication tool for the hearing-impaired community, and its accurate recognition is vital for enhancing human-computer interaction. This paper proposes an improved YOLOv10n sign language recognition model (YOLOv10n-ESHM) with three key optimizations: 1) Introducing the EfficientNet-MBConv module (C2f-EMBC) into the C2f module to enhance feature extraction while reducing computational overhead; 2) Incorporating a Self-supervised Equivariant Attention Mechanism (SEAM) at the detection head to improve robustness against complex backgrounds, occlusions, and lighting variations; 3) Adopting the Inner-MPDIoU loss function to optimize the alignment between predicted and ground-truth bounding boxes. Experimental results demonstrate that the improved model achieves a mean Average Precision (mAP) of 96.9% on the American Sign Language Letters dataset, representing a 2.4% increase over the original YOLOv10n. Additionally, computational complexity (FLOPs) is reduced by 13.1%, and the detection frame rate (FPS) improves by 21.4%, significantly enhancing both accuracy and efficiency in sign language recognition.